►
From YouTube: Ceph Performance Meeting 2022-06-30
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
I
don't
think
we're.
Gonna
have
a
whole
lot
of
pull
request
stuff
to
go
over
here.
It
looks
like
things
are
pretty
quiet
right.
A
Now
I
only
saw
one
new
pull
request
and
actually
pc.
I
might
ask
you
about
this.
It
looks
like
there's
this
new
rgw
atomic
writer
benchmarking
component
being
written
any
any
interesting
things
going
on
there.
B
A
Have
you
guys
seen
anything
interesting
yet
with
it.
B
I
have
no
idea,
I
haven't
really
looked
at
this
one.
I
just
know
that
it
started
fairly
recently
cool.
A
All
right,
well,
yeah
I'll,
be
very
interested
to
hear
what
you
guys
find
with
it
exciting
all
right
well,
so
that
was
the
only
thing
I
saw
over
the
last
like
week
in
terms
of
of
new
pr's
that
were
really
performance
related
and
not
much
is
happening
with
closed
or
updated
either,
maybe
because
of
the
the
upcoming
holiday
in
the
us,
possibly
so
yeah
that
that's
all
I
really
had
there
was
there
anything
I
missed
from
anyone.
A
All
right
well,
then,
in
terms
of
discussion
topics
I
also
don't
have
a
whole
lot.
Since
last
week
I've
been
working
on
basically
a
new
blog
post
for
roxdb
tuning
there's
an
existing
one
there
that
actually
well,
maybe
maybe
this
is
a
decent
form
to
bring
it
up.
A
This
is
the
the
post,
that's
on
blog
article
right
now,
and
I
remember
being
kind
of
angry
about
this
getting
posted
at
the
time,
because
it
didn't
really
explain
their
results
very
well
that
they
saw
and
and
maybe
more
importantly,
they're
changing
a
bunch
of
things.
You
know
the
the
memory
auto
tuning
is
is
whatever
fine.
A
That's
not
really
that
big
of
a
deal,
but
if
you
kind
of
scroll
down
towards
the
bottom
they're
changing
the
osd
pg
log
entries
to
like
10
and
dupes
track
to
10,
and
they
they
basically
just
say
that
they
noticed
that
reduced
right
amplification.
So
they
did
it,
which
it's
not
great,
right
mark
when
you
say
they
who
who
are
they?
A
A
A
So
this
has
been
there
for
like
three
years
and
from
what
I
recall
there
were
some
issues
with
people
trying
to
use
these
settings,
which
does
not
surprise
me
in
the
the
slightest,
but
I
guess
what
I'm
kind
of
concerned
about
is
that
some
of
this
is
gone
out
in
the
wild.
It
sounds
like
so
this
new
post
that
I'm
writing
for
raxdb
tuning.
A
It
kind
of
looks
at
some
of
these
settings
and
what
they
do
and
tries
to
kind
of
describe.
You
know
what
what
settings
are,
what
and
maybe
offer
some
new
updated
ones,
but
I'm
also
going
to
be
really
careful
in
it
to
say,
none
of
this
has
gone
through
our
qa,
so
you
know
use
this
at
your
own
risk,
but
you
know
here's
how
to
understand
what
all
these
things
do.
A
I
wasn't
intending
in
that
article,
though,
to
talk
about
things
like
pg
log
or
the
memory
auto
tuning
or
anything
like
that.
So
I'm
kind
of
wondering
you
know
do
people
think
this
is
a
big
deal.
Is
this
something
that
we
need
to
address?
Or
is
it
good
enough
to
just
kind
of
you
know
let
this
just
kind
of
fade
away?
It's
pretty
old!
Now.
E
You're,
a
lot
more
likely
to
have
them
show
up
in
the
same
in
the
same
short
log
section,
so
it
doesn't
actually
ever
make
it
into
rock's
tv.
A
A
E
D
E
D
Yeah,
so
for
the
pg
log,
when
you
go
from
100
to
10
it
means
you
trim
them
much
earlier,
which
I
don't
know
what
impact
should
it
have
it's
going
to
be
more
expensive
because
the
writer
had
log
update
going
to
be
10
times
as
expensive,
but
streaming
is
streaming.
I
I
don't
see.
What's
the
benefit,
it's
not
that.
D
A
Because
then,
presumably
right,
everything
that
we've
seen
is
is
that
that
should
make
it
so
that
when
you
do
the
flush
you
you
may
not
have
to
actually
do
the
right.
You
just
do
the
tombstone
correct,
because
if
roxcb
won't
know
that
it's
not
already
there
in
the
database,
they'll
have
to
persist
the
tombstone,
I
think,
but
it
should
be
able
to
avoid
the
initial
right
because
it
can
just
get
rid
of
it,
because
it's
gone
from
its
perspective
right
one
less
right.
A
Greg
you
would
still
have
to
do
the
the
tombstone,
wouldn't
it
because
it
doesn't
know
at
that
point
that
the
entry
doesn't
already
exist
in
the
database.
E
E
A
B
D
F
D
But
it's
not
going
to
be
any
other
level
if,
in
fact,
the
compaction
happens
when
the
data
is
moved
to
file
and
the
compaction
eliminates
the
data
and
just
put
the
tombstone,
then
level
0
should
only
contain
tombstones.
So
there's
nothing
to
go
down
because
it's
just
two
stones
and
if
the,
if
the
only
level
you
got
is
level
zero,
then
once
the
heat
level
zero,
you
remove
them
because
there's
nothing
there.
A
A
But
the
the
downside
of
doing
this
right
now
is
you
have
very
few
pq
log
entries
like
your
it's
going
to
make
recovery
slower
and,
like
you
have
very
few
dupes
tract,
which
you
know
is
like
it
kind
of
almost
counter
to
it's
almost
like
you
just
don't
even
have
it
right
I
mean
not
exactly,
but
there
are.
There
are
definitely
trade-offs
to
doing.
E
E
E
A
A
A
A
I
think
some
of
the
performance
conclusions-
this
were
a
little
crazy
too,
but
I
guess
that's
also
kind
of
beside
the
point
of
this.
So.
A
I'll
I'll
probably
have
the
the
roxdb
tuning
thing
done
either
later
this
week
or
beginning
of
next
week.
So
hopefully,
then
people
in
this
group
can
take
a
look
at
it
too
and
before
we
actually
post
it-
and
I
can
I'll
try
to
see
if
we
can
get
some
feedback
on
before
it
goes-
live.
A
Oh
josh
hi.
It's
definitely
not
fine,
but
thanks
for
tuning
in.
A
All
right
well,
so
this
is
basically
all
I've
got
right
now,
I'm
still
working
on
looking
at
the
osd
threading
behavior
and
there's
some
interesting
stuff
going
on
there,
but
this
other
article
kind
of
ended
up
taking
precedence,
so
I
won't
probably
get
that
until
back
to
that
work
until
next
week
at
the
earliest,
and
that
was
basically
all
I've
got
right
now.
A
If
not,
then
I
guess
we'll
we'll
wrap
up
early
and
for
folks
in
the
us
have
a
great
holiday
weekend
and
for
other
folks
have
a
great
weekend
in
general.